Intro
Contents
At Alpha Audio, we dare to claim that we have a lot of insight into how to achieve optimal networking for music playback. We have done a lot of listening tests, set up comparative tests and taken measurements to understand what is happening. In this background article, we lay out this knowledge for you.
Ever since music streaming has taken a dominant role in the hi-fi hobby, there have been discussions about whether or not the (data) network that transports the audio bits and bytes has an effect on quality. Even we at Alpha Audio were quite skeptical at first, since a data network has error correction. So how is it then possible that differences are audible? After all: when data corruption occurs, you just hear a tick, click or a dropout.
Initial tests
However, Alpha Audio has always kept an open mind when it comes to trying and investigating interesting propositions and phenomena. So also the influence of network tweaks.
Our assumption was (and is): if there is a difference in switches and network cabling, then the difference between a standard UTP cable and fiber must be very large…. To that end, we purchased a fiber to UTP converter and set up a simple A / B test. The difference…? VERY audible! Oy! … That test was the end of our unconscious assumption that there could be no difference. However, the bits and bytes were the same. So how could that be? The only difference between the two is the transport medium…. And what then remains is (especially common mode) noise. After all, fiber is the ultimate isolation.
Noise measurements
The assumption that it must be in (common mode) noise led to the first major test of switches and power supplies of switches. The first, large background article can be found here. And the first major test we did here.
The conclusion: there is quite a difference in switches and their noise levels. The power supplies of the switches also differ greatly. And swapping out the standard power supplies for a decent (quiet) model, lowers the noise floor significantly.
Jitter measurements
The next step we wanted to do was to measure jitter. That was another serious challenge, because not only do you need expensive measurement equipment; the environment in which you measure also plays a big role.
We managed to make considerable strides there, thanks to the help of engineers with whom we have good contact. As a result, we started working with active probes and made an RF measurement box. All to be able to measure more precisely.
Again, the results speak for themselves: swapping out a switch lowers the noise floor and thus the jitter on the clock that drives the dac. These are reproducible measurements and results that have also proved audible.
We were not only able to gain insight into the influence of switches: cables also have an influence, although these differences are usually smaller. But anyway: every little bit helps….
Other issues
That hardware has an influence is pretty obvious to us. But what about software? Consider streaming software and settings of switches and routers.
That too has an influence. The differences between J River, Roon and Audirvana, for example, are quite audible. You can read about that in this article and in this latest background piece.
And switch and router settings also have an impact. Think flow control, priorities and Vlans that can isolate and prioritize traffic (by setting the QoS priority per Vlan). We use PF Sense to keep our network running smoothly.
In this article
The following pages detail our findings by network component. We work from the router to the switch, power supplies and cables. For each component, we outline our insights.
Firstly, fantastic research and very helpful to all.
I do have a few questions as I am still in the early stages of creating my ideal audiophile network. At present my network is fibre to premises, Ont converted to Ethernet, to link router, cat 6a shielded cable to nether gs105e switch, cat 7 Audioquest pearl shielded cable, Nordost qnet with smps, Audioquest Cinnamon to Tambaqui dac.
Before the Cinnamon I had all cable shields bonded to telegartner connectors. But I found that when I touched the qnet switch and dac I was getting a strong tingling. I tried disconnecting the shield from the cable feeding the qnet which helped a bit. I then tried the cinnamon between dac and qnet, which reduced the tingling a bit more.
With all Ethernet cables removed from the qnet there was a strong tingling when touch the dac and Qnet because of the ground difference.
All of my switches use smps so none are conn cited to ground and therefore the shields are not connected to ground.
I checked with Audioquest and their network cables the shield is on connected at the receiving end (in my case the dac end).
Is the difference in grounded(dac) and ungrounded Qnet a problem? Should my cables be shielded end to end? Should I use a switch with an earth so that the cable shields are grounded?
Any thoughts or suggestions would be greatly appreciated.
Kind regards to all. Richard W
First off, the Nordost QNet with an SMPS can be significantly improved by upgrading to an LPS (or if not possible by shunting the SMPS). Even if you’re using a Telegärtner or another type of metal connector on your network cable, it doesn’t necessarily mean it’s connected to the shield—it might be, but not always. Cat 8 spec could be an indication.
When it comes to grounding, there’s often room for improvement. However, grounding can be a complex topic. It’s worth reading up on it, particularly from manufacturers who specialize in grounding solutions for hi-fi systems.
Finally, try connecting the AudioQuest Pearl as the last cable in your setup instead of the Cinnamon. If you don’t notice an improvement, it could indicate that something else in your system needs attention. Remember, the final meter before the streamer (or DAC) is where your best gear should go. But go back and evaluate your network all the way to the router, for example also replace the standard SMPS for the Netgear as discussed by Alpha.
Sorry, my mistake, I read Audioquest Diamond. It is a pearl. Nevertheless try with much better shielded ethernet cables, especially the last one.
Hello Richard. It is certainly not wanted that you feel the current when connecting your DAC through an Ethernet cable. It sounds like a faulty Ethernet port on the DAC side.
Hello Martin,
If I remove the ethernet cable from the dac I still get a tingling (current) when touching the DAC and Qnet switch at the same time. Does this occur with the switches you have tested?
No. And since you feel the current on both devices when they are unconnected, I think you use a power distribution block in which both devices are plugged in?
If so, I’d take a hard look at that power distribution block, it seems there’s current leakage somewhere.
I also assume that the power distribution block or wall sockets are earthed.
If that’s the case, unplug everything, and plug in one device after another to figure out where the leakage is originating.
I agree. 🙂
Thank you Martijn and Paul,
Yes dac and qnet both connected to Audioquest 5000 distribution unit all grounded (I checked with a socket tester).
I will check each unit separately but using multimeter the Vitus amp and dac cases do not have current. As the qnet smps does not have an earth i presume the Qnet has a floating ground and perhaps when I touch Qnet and dac the current in the Qnet is just seeking the lower ground. Also perhaps your network is earthed at some point. I am also asking Nordost with regard to shielding and grounding and will post anything useful. So far I have learnt that Valhalla and Helmsdale network cable shields are bonded both ends which blue heaven shield is bonded only at one end.
Jaap,
Thanks for all the work you did here!
First off why in some areas is Phase Noise depicted as PS? It would always be in dB.
Now from digital audio we know the better (negative) the Phase Noise is below 10Hz the less effect that jitter has on the SPDIF, I2S and other audio protocol formats.
The big question here is…. Is that the same for Ethernet and or types of Ethernet? I was heavily involved with the design of 100M Ethernet when I was designing bridges and routers back in the late 80’s early 90’s. One of the reason’s I prefer 10/100 over 1G and above is the full duplex protocol. There are so many less variables to consider. When you hit up 1G and above and the protocol becomes nibble (4bits) 1/2 duplex then there is so much more going on and Phase Noise and Jitter may not account for as much as how the receiver processes each bit.
Thought?
Thanks,
Gordon
Hey Gordon,
Ik recalculated the phase noise to ps for ease of reading. You can see the graphs in the article. I also did not measure phase noise on the network, but on the clock in the streamer. The closer to the carrier, the more critical. So phase noise <10 Hz has more audible impact. You can also clearly see that when you recalcucate it. High phase noise at 1 Hz = high jitter (ps).
Jaap,
Most of the switches use a 25Mhz or 50Mhz crystal with an internal generated clock. Now the big problem here is what the part is doing. On the receive side it’s actually jacking that clock around in a PLL which will create more jitter. Another fundamental problem with most of the testing I have seen is the time period. We did a test on a symetricon that we have with some of the most expensive oscillators. The difference in phase noise with the units powered on for 10 minutes as compared to 24 hours was crazy. Also remember to tell people by high phase noise you mean less negative. So -110dB is much much better than -60dB. A bigger deal than jitter on any of these parts is signal integrity. A receiver will sample each edge x8 times. That determines the signal quality that you can read from most switch parts. The average over that 8x times determines a 1 from a 0. But really who cares if the protocol is error correcting. Well!!! actually sending and resending error packets leads to all kinds of noise in the system. Same with 10/100 being less noisy that 1G and above. With 1G and above being 1/2 duplex the amount of noise in switching directions is drastically higher than something that is full duplex.
Hi Gordon,
”We did a test (…)” did you publish the results somewhere?
Regarding clock stability and warm-up I can not agree more; ensure your switch is powered on for at least a couple hours before critical listening. This allows the oscillator to stabilize, reducing phase noise.
I think you are also saying when choosing switches, prioritize ones with lower phase noise. Sorry to repeat my self in these pages but ”audiophile” switches are engineered to minimize clock phase noise (check the specs) please try.
Avoid resends from error correction by using high-quality (I am not saying expensive) Ethernet cables to minimize errors and prevent retransmissions, which can introduce system noise. You can succesfully use unshielded UTP cable in your (largest) upstream section.
Ensure a clean network environment, not only the last part close to streamer or dac, avoiding excessive interference (e.g., from power lines or other electronic devices).
Use 10/100 Mbps Instead of 1G or Higher, is worth comparing but sometimes not so practical when copying files and libraries over the same connection.
Thank you for the reply Paul,
I have upgraded the SMPS on the netgear and router to IFI’s and intend to upgrade the Qnet to a Farad Super6 once i have sorted out the issue with grounding. The tingling sensation when touching the dac and Qnet simultaneously is quite alarming. Does anyone else’s dac and switches do this?
I have checked the shielding on the cables and it is all connected (bonded) to the connectors.
With shielding connected end to end the tingling is at its worst. If I use cables with shields connected only at one end the tingling reduces but is still present which means that the ground path is via the signal wires, which i presume is not good.
Enhancing Streaming Audio with Optimized Switches
Alpha Audio’s recent reference to their article as an “absolute and ultimate guide” to streaming audio networks is ambitious and reflects the team’s dedication to this topic. Their detailed work, particularly on the role of the final switch and the impact of a high-quality linear power supply (LPS), offers valuable insights to the community. However, it’s important to recognize that there is still more to explore before reaching an “absolute” or “ultimate” understanding of streaming audio network optimization.
For example, Alpha Audio acknowledges better sound quality when streams are separated via VLANs on the router, yet they dismiss the possibility of hearing upstream improvements beyond the last switch. Additionally, their tests focus primarily on standard switches, which—while effective when paired with an excellent LPS—cannot match the intentional noise management and design benefits of audiophile-grade switches. By omitting these, the guide misses an opportunity to address the broader potential for network optimization.
True progress lies in addressing noise at every stage of the network, including routers, intermediate switches, power supplies, and cables. Many audiophiles report audible improvements in these areas, even when measurable differences are elusive. Alpha Audio’s current approach, while thoughtful, risks oversimplifying the nuanced interplay of components within a high-performance network.
Lastly, breaking the noise chain between the switch and streamer is crucial for maximizing sound quality. High-quality Ethernet cables, designed with proper shielding and grounding strategies, play an essential role in this process, preventing noise propagation and enhancing the clarity of the audio signal.
Alpha Audio has done commendable work in pushing the conversation forward, but there remains significant room to expand the scope of exploration. By embracing these additional factors, they could further solidify their leadership in streaming audio network optimization.
This comment has used ChatGPT, or a tool which is alike, to write the content. The author has confirmed this.
It is obviously impossible to prohibit the use of these tools, it is like trying to shout at the sea that it has to retract when it is edging towards high tide.
But there are a few rules that do apply, and which we will amend in our Do’s and Don’ts ( https://www.alpha-audio.net/dos-and-donts/ ):
– Above text is potentially confusing, since it is written in a third person form.
– It is therefore unclear if any conclusion or opinion expressed is generated by the AI tool or the person. It now reads as a summary of everything we have ever written on the subject, being thrown into generative AI and now the bot is telling us were the flaws are.
So, in the future, we will remove any third person / summary like post without discussing it first with the author.
I understand that my posts may have given the impression of being AI-generated, but I want to clarify that I only use AI to refine my writing for clarity and correctness, not to generate the content itself. The ideas, insights, and solutions I share are entirely my own, reflecting my authentic thoughts and efforts. I deeply value the open exchange of ideas here at Alpha in this forum and believe that my contributions helps a more informed and balanced discussion about sound quality.
When I write about Alpha Audio, I am referring to you, Martijn and Jaap, as you/they share your opinions here. I cannot always know who specifically wrote a piece, but my intent is not to discourage you. While we may not always agree, I hope to present a broader perspective that can enrich the discussion.
That said, it feels like more focus is being placed on enforcing the rules than engaging with the content itself, which can be disheartening. I genuinely appreciate the good work you’re doing and hope we can keep the focus on constructive dialogue.
“For example, Alpha Audio acknowledges better sound quality when streams are separated via VLANs on the router, yet they dismiss the possibility of hearing upstream improvements beyond the last switch.”
Creating Vlans can optimize the datastream and prioritize the medianetwork above – for example – a guest network of smart home network. That can prevent drop-outs. But can also improve the flow of data and prevent ‘bursts’ in traffic. We haven’t done any in-depth research on this topic, so I cannot offer you any more than: I heard a difference / improvement in sound quality. Just like optimizing settings in switches.
This is completely different from buying better / expensive switches or power supplies. I also don’t believe that creating vlans or optimizing settings in switches lowers the noise floor. It just optimizes the datastream and maybe with that lowers traffic on the media-network (and maybe with that lowers noise, but we have to dig deeper for that).
Could you please confirm if you often are using a NAS/PC upstream in your signal chain, when you do your listening tests?
I ask since i have a feeling that your upstream components (router) are polluted due to this(?) and therefore we see this differences in views on this topic. I don´t have any computing device directly connected to my router. (don´t have a PC/NAS at all connected)
Some of us don´t understand why your view is drastically different when it comes to upstream improvements and it would be great if you could shed some light if an upstream NAS/PC has such a big impact, as i now think they might have.
Yes, I have a NAS connected to the network (seprate switch that is connected to the router). Where else would I have stored the flac-files? I don’t like having the files on a drive in the audio-pc, for that generates noise as well. I rather have it on a nas, far away from my audiosystem.
Thank you Jaap for sharing this!
I actually think this matters quite a lot.
It’s all about minimizing RFI noise that comes with the digital signal so it doesn’t mess with my DAC. These days, when I’m able to control and reduce the noise, the location of the files—whether on a local drive, NAS, or streaming service—matters a lot less. That said, I’ve found that local SSDs in the streamer often work best, since they can avoid network-induced noise more effectively.
A switch doesn’t pass on RFI noise that’s already reached it, unless that noise can travel through the shielding of a cable. The audiophile switch is designed to take advantage of this and further improve the system by reducing the noise the switch circuitry itself produces. It also helps stop RFI from getting into the switch through things like holes in the case or unused Ethernet ports. In the end, I use a switch or an optical link not to preserve the digital signal’s integrity at the streamer’s input, but to reduce the RFI noise that makes its way to the analog components downstream, where it can mess with the sound quality.
Thanks for this article. Very timely for me, as I’m about to add a separate streamer, something beyond an SSD plugged into an Oppo.
Is this article available as a PDF? Other than convenience, trying to print page by page is hindered by the Alpha–Audio banner at the top of each page, sometimes blocking text. Thanks,
You need to compare the sound quality when using WiFi instead of cabling (whether with fiber or not).
The WiFi radio in streamers are often not shielded well enough, known exception is Auralic, and introduce more problems than they potentially solve.
Very interesting – however I cannot help wondering what exactly the or rather where the benefits to the sound arise. Is it due to the media conversion or the galvanic separation?
Would be interesting to compare to a streamer/dac with galvanic separation.
To me one fundamental flaw with Ethernet is that was based on a collision system where random timers on the network adapters sends the signal with varying timing to avoid collision. Yes, now we have switches so no collision but there is no timing signal in Ethernet and never will be. So, the way the signal is assembled and buffered is what determines the sound quality.
This is also why when you stream Apple Music using Apple lossless then there is a 2 second delay to allow for a buffer to be compensating for network errors and delay. As far as I know it is the only protocol with built in buffer. Yes, you can add it to all forms of streams but Apple included it from the start.
Anyway, just my question and observation.
https://www.alpha-audio.net/background/how-a-network-switch-affects-audio-playback-an-extreme-deep-dive/
“…however I cannot help wondering what exactly the or rather where the benefits to the sound arise.” and “So, the way the signal is assembled and buffered is what determines the sound quality.”:
It seems there is no really perfect buffer for digital streams. There’s all sorts of re-clocking, super-femto whatsoever clocks, PLLs, … but in the end none of these things seems to be able to reduce slight and cyclic variations (aka jitter) in data timing at the DAC to exactly zero.
Put another way: everything that influences the quality of the digital stream in from of the final DAC stage has potentially influence on sound quality.
At least that’s my layman’s understanding, or should I say… belief! 😉
I am surprised that @anders and you keep focussing on the digital stream. I explicitly explained that the issue is not the data, nor the stream. I already checked that numerous times.
I explained in this article – and others – that the issue is the electrical noise that impacts the clock in the dac or streamer. If this clock stability gets impacted (jitter), the result is obvious: conversion errors / distortion. It’s not very complicated and actually very logical.
… honestly I think we’re on the same page.
For me the focus on the term “electrical” noise leaves me wondering why then an optical TOSlink connection from streamer to DAC is not the bullet proof solution – from my experience it is not. And you point out in your articles that the resulting jitter is the culprit of sound quality losses on the digital side
Hence I think of “electrical noise” as “timing variations” / aka jitter in the digital data stream. Maybe too philosophical…
Anyway. Thank you again for all the hard work!
TOSlink has many other known issues.
Toslink has other issues, like Martijn says.
First of all: it works as an spdif-interface. Spdif is a completely different protocol than TPC/IP. TCP/IP is package based, spdif isn’t (serial data).
Second of all: there are two extra conversions necessary for toslink. With spdif (no error correction!) that is not ideal. Rather use coaxial, or better: AES (balanced) for that will eliminate noise (common mode).
I can go on, but in my opinion these are the most important. If you need to use toslink, invest in a very good cable… most cheap toslink cables are crap, and that will result in diffraction, which will destroy everything.
Jaap, well… seems you have a point. A very strong point.
“I explained in this article – and others – that the issue is the electrical noise that impacts the clock in the dac or streamer.”
I still have an Auralic linear PSU (15 Volts instead of 12V, but what the heck…) lying around.
Powering the “last switch” before the streamer with it instantly cleans up the sound considerably.
Replacing the stock fuse (that came with the Auralic PSU) with a HiFi-tuning variety even more. It’s… dramatic.
And it’s only the PSU of the switch.
Anyway. Thanks again (seriously) for your guidance. Cheers!
Thank you! The English term for what you are experiencing with those objections is “cognitive dissonance.”
You guys are working hard! Amazing work in this area and a great help for the community!
We have recently discussed this, and i am sure you are going to get simial question from others, so i am thinking it might be good for you to explain your thoughts on this a bit deeper.
I am not the only one that is reporting rather big impact also when doing changes prior to the switch. In fact, in my own experience that was a big wake-up call for me where i realized the scale of this problem, even more. This of course gets more noticeable the lower noise floor you have later in the signal chain (in my experience and that also makes sense in my head)
In my case, i used to have the GS105 switch + Ifi Ipower X with a wireworld (the red one) cable to my streamer. Then later i also implemented a Ifi Ipower X on my router and a used long AQ Vodka cable between my switch and router. I put the more expensive cable there since it happened to be longer (5 meters) and i needed the length there.
It was a very big and instant improvement, that was impossible to unhear.
Now, since you don´t have the same view on this (in general at least) and can´t go back and test this again, since my setup now is different, i have given this some thought to try to understand why i had this experience.
1. First of all, i live in an apartment with 20-30 WiFi connections coming up on my phone. I know WiFi radiation has an impact on the cables, but could this explain this big difference?
2. My router (and switch) was only physically connected to the switch->streamer, meaning that it was not contaminated by anything else, if that matters? (i have a feeling many have some sort of NAS or similar there…?)
3. My router and switch was on the same power line. My HiFi setup was on a separate power line, if that matters.
I should say that i do understand that the Vodka cable is rather expensive and maybe that money is better served somewhere else! But still, the logic seem to be there (for me at least), and maybe there are cheaper ethernet cables that do a similar job?
I hope i don´t come across as complaining since i really just would like to understand why your findings are very different than mine (and many others on HiFi forums, it seems).
If you hear a difference, you hear a difference. I never experienced any differences at home, or at the office. I also couldn’t measure differences when I setup something similar in the lab. Sorry.
I have two more question, on how you have tested this, if that is ok?
1. When you tried with better ps on the router did you use “audiophile” grade ethernet cables also from the router?
If you did:
2. Do you have a lot of pollution into your router, like NAS and other noisy things connected?
Thank you so much for answering!
… I noticed that my NetGear Switch (GS108Ev4) influences the sound depending on the connection speed. Read: When I manually set it to 100Mbit/s, it sounds worse. The default 1Gb/s sounds better.
Then I also noticed that an older long Ethernet cable I used up till recently seems bad / of worse quality, since it only allowed other components to connect with 100Mb/s instead of 1Gb/s.
Maybe the long ethernet cable that you used before the AQ Vodka influenced the connect speed into the switch? Even if not, after learning about all these funny effects of noise and so on I am not surprised at all, any more. Hearing is believing. 😉
Tobias,
In your case the Vodka cable replacing the Wireworld cable is implementing a proper CAT7 twisted pair design instead of the parallel twinax geometry of the Wireworld cable. Esspecially on a medium/long lenght of 5 meter, I’m not surprised that is yielded better results. Purely based on this I wouldn’t question the conclusion of Alpha that the biggest impact on network optimization is in the last few inches (last switch and ethernet cable).
However, as the genius Alpha guys also mentioned, tuning your network end-to-end does make a certain amount of sense. Maybe the 80/20 rule is applicable. But even if optimizing the first 80% of your network section only yields just 10% quality improvement, in a well balanced high quality hifi set that might.judt be the cherry on the cake.
Things to consider:
1) upgrading/tweaking your router yields minimal benefits. Setting up vlans and splitting your domestic network also physically from you audio network does.
2) Decoupling your router and switches with dampers/absorption platforms to minimize vibrations will give improvements
3) same as your point 3, power lines doesn’t have to be exactly the same, as long as you also apply the right polarity, same as with your hifi rig. In the end it is all modulated power,, so ensure you use the same polarity end to end.
4) using good quality ethernet cable early in the network does help a bit. I shifted all my obsolete LAN cables after upgrades ‘upstream’ and it at least didn’t hurt anything. Even upgraded the coax cable between router and 1st switch for a nice design assembled by Peacock Audio. May be confirmation bias to justify the 150 euro’s spend, but I did sense some more blackness and piece (of mind?).
5) apply a proper shielded twisted pair ethernet cable between tonyour audio switch. I chose Audioquest Carbon.
Still the biggest improvement, as concluded by Alpha Audio as well, is your last switch and cable to the streamer. I’m still surprised how much impact that makes. Three years ago I found spending serious money on that questionable, to say the least. Yesterday I tried a 3.5k power cable on my switch alone, and I was blown away with what I heard. Quickly disconnected it, to prevent an unsuppressable itch…. 😉
I agree that the biggest difference is the last switch (ps) and cable, no question about that!
I didn´t replace the Wireworld with the Vodka on the 5 meter from my router. There i replaced a Supra cable. The Wireworld was still my last cable.
I just try to understand why it had such a profound impact in my setup, to implement the Vodka and PS from the router and not noticed at all by Alpha Audio. But there are many variables of course so I also understand and respect that they have to draw the line somewhere on what they can investigate in this area…
To support my theory logically, if that cheap switch is so good at stopping the noise prior why can´t we stop this noise in the streamer, or in the DAC step? In my experience you can take the noise reduction much further if you make sure to feed that switch with a less polluted signal.
This lead me to a completely different solution actually since i noticed that you ideally don´t want Power Supplies at all in the network signal. It is possible to reduce the number of PS to one if thinking very differently.
Tobias, Sorry I didn’t understand your setup or made wrong assumptions. Tried to figure out where the differences you heard could originate from.
piece (of mind?) = peace (of mind?)
and excuse some other typos.
Would be nice if the platform allows editing messages to correct it.